Method __init__

model/classification/classifier.py:26–90

(self, dataset, config)

class Classifier(torch.nn.Module):
    def __init__(self, dataset, config):
        super(Classifier, self).__init__()
        self.config = config
        assert len(self.config.feature.feature_names) == 1
        assert self.config.feature.feature_names[0] == "token" or \
               self.config.feature.feature_names[0] == "char"
        if config.embedding.type == EmbeddingType.EMBEDDING:
            self.token_embedding = \
                Embedding(dataset.token_map, config.embedding.dimension,
                          cDataset.DOC_TOKEN, config, dataset.VOCAB_PADDING,
                          pretrained_embedding_file=
                          config.feature.token_pretrained_file,
                          mode=EmbeddingProcessType.FLAT,
                          dropout=self.config.embedding.dropout,
                          init_type=self.config.embedding.initializer,
                          low=-self.config.embedding.uniform_bound,
                          high=self.config.embedding.uniform_bound,
                          std=self.config.embedding.random_stddev,
                          fan_mode=self.config.embedding.fan_mode,
                          activation_type=ActivationType.NONE,
                          model_mode=dataset.model_mode)
            self.char_embedding = \
                Embedding(dataset.char_map, config.embedding.dimension,
                          cDataset.DOC_CHAR, config, dataset.VOCAB_PADDING,
                          mode=EmbeddingProcessType.FLAT,
                          dropout=self.config.embedding.dropout,
                          init_type=self.config.embedding.initializer,
                          low=-self.config.embedding.uniform_bound,
                          high=self.config.embedding.uniform_bound,
                          std=self.config.embedding.random_stddev,
                          fan_mode=self.config.embedding.fan_mode,
                          activation_type=ActivationType.NONE,
                          model_mode=dataset.model_mode)
        elif config.embedding.type == EmbeddingType.REGION_EMBEDDING:
            self.token_embedding = RegionEmbeddingLayer(
                dataset.token_map, config.embedding.dimension,
                config.embedding.region_size, cDataset.DOC_TOKEN, config,
                padding=dataset.VOCAB_PADDING,
                pretrained_embedding_file=
                config.feature.token_pretrained_file,
                dropout=self.config.embedding.dropout,
                init_type=self.config.embedding.initializer,
                low=-self.config.embedding.uniform_bound,
                high=self.config.embedding.uniform_bound,
                std=self.config.embedding.random_stddev,
                fan_mode=self.config.embedding.fan_mode,
                model_mode=dataset.model_mode,
                region_embedding_type=config.embedding.region_embedding_type)

            self.char_embedding = RegionEmbeddingLayer(
                dataset.char_map, config.embedding.dimension,
                config.embedding.region_size, cDataset.DOC_CHAR, config,
                padding=dataset.VOCAB_PADDING,
                dropout=self.config.embedding.dropout,
                init_type=self.config.embedding.initializer,
                low=-self.config.embedding.uniform_bound,
                high=self.config.embedding.uniform_bound,
                std=self.config.embedding.random_stddev,